Connect Cituna to Claude with MCP

MCP, the Model Context Protocol, lets Claude call tools on your behalf. Connecting Cituna means you can ask "why is ChatGPT not citing us?" and Claude reads your real scans, your Google Search Console data and your citation sources to answer, instead of guessing from what it remembers about your brand. The hosted connector is the quickest route: paste one URL, sign in to Cituna in your browser, approve access. Nothing is installed, so it works in Claude on the web and on mobile as well as in Claude Desktop and Claude Code. If you would rather run the server yourself for scripts or CI, it is also on npm as cituna-mcp and takes one line of config plus an API key.

1. Add the connector URL

In your MCP client, add a custom connector and paste https://mcp.cituna.com/mcp as the URL. In Claude that is Settings → Connectors → Add custom connector.

There is no config file to edit and no API key to copy for this route.

Connector URL

https://mcp.cituna.com/mcp

2. Sign in and approve access

Your browser opens a Cituna sign-in. Approve the access request and the connection completes. Authorisation is per workspace, and you can revoke it from Integrations → Claude / MCP access at any time.

That is the whole setup. Claude will now list Cituna tools alongside its own.

3. Ask a first question

Start with something that forces Claude to read real data rather than answer from memory: "Which of our tracked prompts do no engines cite us on?" or "Where do we rank on Google but stay invisible in AI answers?"

A good first session establishes the shape of the workspace: which brands are tracked, whether Search Console is connected, and what the last scan measured.

4. Optional: run the server locally for scripts and CI

For automated jobs, install the published package instead. Generate an API key under Integrations → Claude / MCP access. It starts with cituna_sk_ and is shown once. Then add one of the configurations below.

The hosted connector and a local install expose the same tools; the local route simply removes the browser sign-in, which is what a CI job needs.

Claude Code

claude mcp add cituna --scope user \
  --env CITUNA_API_KEY=cituna_sk_your_key \
  -- npx -y cituna-mcp

5. Optional: Claude Desktop config

Claude Desktop reads a JSON config. Add the cituna entry, restart Claude, and the tools appear.

claude_desktop_config.json

{
  "mcpServers": {
    "cituna": {
      "command": "npx",
      "args": ["-y", "cituna-mcp"],
      "env": { "CITUNA_API_KEY": "cituna_sk_your_key" }
    }
  }
}

Troubleshooting

Claude says my API key is invalid or expired.

On a local install this is almost never the key. It is usually a stale build of the package, or a redirect between http and https stripping the Authorization header. Re-run with npx -y so you get the current version, and make sure any proxy in front of the request forwards headers on redirect.

Which tools can actually change things?

Read tools are available from Starter. Write tools, meaning queueing an article, marking one published, changing a gap status or starting a scan, need Pro or Max. The connector reports the plan when it starts, so Claude knows what it may attempt before it tries.

Claude answers about my brand without calling any tools.

Ask a question only your data can answer, such as your citation rate on a specific tracked prompt over the last week. Vague prompts like "how is our AI visibility" let a model fall back on general knowledge; naming a metric and a period forces a tool call.

Does connecting give Claude access to my Google account?

No. Cituna holds the Search Console connection and exposes summarised results through its own tools. Disconnecting Search Console in Cituna cuts off that data to Claude immediately, and revoking the MCP grant cuts off everything without touching Google.